Papers

The dynamical formation of a complete Bose-Einstein condensate and the asymptotic purification of trapped bosons coupled to a quantized radiation field are rigorously established. By tracking the macroscopic one-particle density matrix $\kappa_T$ through a nonlinear resonance cascade, it is demonstrated that radiative relaxation irreversibly depletes all excited trap modes, even when the initial macroscopic state $\kappa_0$ is genuinely mixed. Here $\kappa_T$ is an average, with respect to a de Finetti measure $\mu_0$, over mean-field particle-field orbits, so that the mixedness of $\kappa_0$ reflects a statistical ensemble of orbits. Under the assumptions of a non-vanishing initial ground state component and strictly positive downward Fermi-Golden-Rule rates, strong trace-norm convergence to a pure condensate is proven, $$ \lim_{T\to\infty} \| \kappa_T-|\chi_0\rangle\langle\chi_0| \|_{\mathfrak{S}_1(\mathfrak{h})} = 0 \, . $$ Furthermore, the von Neumann entropy of the macroscopic one-particle density matrix is shown to vanish asymptotically, $$ \lim_{T\to\infty}S[\kappa_T]=0 \, , $$ without any claim of monotonicity in $T$. Thus, any initially mixed, positive-entropy state ($0<S[\kappa_0]<\infty$) is purified to the ground state $|\chi_0\rangle\langle\chi_0|$. For the proof, we derive the Hartree hierarchy of the coupled system of bosons and photons from the finite-$N$ BBGKY hierarchy of the microscopic quantum many-body system, establishing trace-norm convergence in the mean-field limit as $N\to\infty$. The limits are taken successively; first $N\to\infty$, then the weak-coupling limit, and finally $T\to\infty$. Although the global particle-field evolution remains unitary, the macroscopic one-particle state converges to a pure state as energy and correlations are continuously radiated into the dispersive field.

Action-conditioned world models can reduce costly real-robot interaction in reinforcement learning for Vision-Language-Action models (VLA-RL), but hallucinations can mislead policy optimization. Extending them to richer VLA interfaces and long-horizon tasks raises challenges in spatiotemporal and cross-modal consistency, exploration-reliability balance in RL, and compounding rollout errors. We present WoVR+, extending WoVR's hallucination-aware WM-RL framework by redesigning the learned environment and imagined interaction protocol: A VLA-interface-aligned world model provides multi-view prediction and proprioceptive updates while mitigating hallucinations. Chunk-wise noise preserves flow-based policy exploration while limiting prediction-degrading action jitter. For long-horizon tasks, stage-local imagined training combines stage-entry resets, stage-specific rewards, local horizons, and controllable sampling to shorten generated prefixes and balance training across stages. WoVR+ improves success over SFT by 21.09% and 16.7% on RoboTwin and 6 real-robot tasks. Long-horizon phone-packaging success rises from 30% to 75%. Videos are available at https://youmo445.github.io/WoVR_Plus.github.io/.

Recently, an analog to the t = 0 moment of our Solar System's meteoritic record was identified for the first time around the Class I protostar HOPS-315. Crystalline silicates were found to recondense from SiO gas that cools as it advects upward from a hot thermostat region in the inner midplane. This t=0 moment may be associated with the formation of the first planetesimals in this disk. We explore whether HOPS-68, a Class I protostar with similar properties and system geometries, is engaged in such refractory recondensation as well. A detection of cooler gas-phase SiO in absorption above a hot disk midplane would allow us to achieve this goal, thanks to a previous detection of crystalline silicates. HOPS-68 was observed with the JWST MIRI and NIRSpec, showing multiple bands of molecular absorption. We use local thermal equilibrium slab models to measure the temperature, column density, velocity, and radial location of these bands. The molecular absorption bands of gaseous SiO, H2O, CO2, and 13CO trace a warm molecular inversion layer (300-500 K) at mid-altitudes above the hot disk midplane from 0.3-16 AU, which is consistently blue-shifted with respect to the system velocity by 33-53 km/s. Atacama Large Millimeter/submillimeter Array (ALMA) observations of HOPS-68 kinematically resolve an SiO-rich jet and wind. We confirm that HOPS-68, like HOPS-315, is undergoing an epoch analogous to the t=0 moment in the Solar System's planet formation timeline. The detected warm molecular layer traces the base of a disk wind, which carries sublimated material from the thermostat region near 2 AU up into the outer disk. As for HOPS-315, the molecular layer in the inner few AU of HOPS-68 is enhanced in bulk oxygen as a result of this sublimation. Since HOPS-68 is less evolved than HOPS-315, this may indicate a prolonged epoch in protostellar evolution in which such oxygen enhancement occurs.

In case a crystal is close to a continuous elastic phase transition, pronounced critical fluctuations emerge. We show that on the level of linear elasticity theory the displacement vector for these critical modes obeys a condition reminiscent of the conformal Killing equation. For an isostructural instability, four critical modes arise corresponding to one dilation and three special conformal transformations. At crystal symmetry-breaking instabilities, the critical fluctuations include a continuum of modes even for crystals with finite size. For certain cases, these comprise the conformal transformations associated with two-dimensional Minkowski space. These results have implications for critical elastic phenomena as well as for extremal mechanical metamaterials.

Close-in sub-Neptunes are born in mean-motion resonant chains that break apart on a timescale of $\sim$100 Myr. If the chains are first ”rattled” — have their free eccentricities excited to a few percent — then breaking occurs on the required timescale. We propose that chains are rattled by internal oscillations tidally driven by the star in the innermost planet. Unlike other proposals, there is no need to invoke any reservoir of planetesimals or external perturbers — the chain unravels from within. We develop a Hamiltonian theory (and an accompanying toy model) which predicts a parametric instability whereby chain librations exponentially amplify when the tidal forcing frequency equals the sum of the internal mode frequency and the chain libration frequency. We also introduce a new eigenmode solver for rotating planets, $\texttt{MAQAM}$, that in combination with the evolutionary code $\texttt{ORCHARD}$ enables us to track the frequency evolution of pressure, gravity, and inertial modes in cooling sub-Neptunes. Inertial modes in the gas envelope and fluid mantle can evolve to become momentarily parametrically resonant. Passage through a single parametric resonance can raise the free eccentricities in the chain to a few percent within a few tens of Myrs.

The creation of cold atoms underpins quantum sensing and precision metrology, computing, timekeeping, and fundamental science experiments. Miniaturizing the lasers and optical infrastructure, which occupy over 80% of the volume of these experiments, through photonic integration could improve robustness, reliability, and manufacturability while reducing cost and enabling portable space-based precision measurements and quantum tools for non-experts. Yet, there has been limited progress towards integration of the laser and optical infrastructure required to generate cold atoms onto a single chip. We demonstrate a photonic engine that generates, controls, and delivers precision light to create cold atoms in a rubidium 3-dimensional magneto-optical trap (3D-MOT). We integrate on a single chip two frequency-agile self-injection-locked lasers, laser frequency and intensity control, passive beam combining and splitting, signal monitoring, and high-efficiency beam delivery to a rubidium vapor cell. The integrated engine is shown to support multiple functions required for generating cold atoms including stable generation of low-noise cooling and repump light, frequency control for atom spectroscopy and stabilization, real-time laser monitoring, routing for laser offset locking, optical power modulation, and large-area three-beam delivery of 3 mW to the physics package. Fully electrically controlled, the engine enables sustained 3D-MOT operation, trapping more than 1 million atoms at an average temperature of 350 uK and supporting atom-number and temperature measurements. These results demonstrate the first single-chip integration of light generation, control, and delivery for cold-atom generation, offering a path toward smaller, more robust, and manufacturable cold-atom systems for applications including cold-atom clocks and sensors.

As generative language models become more capable and widely used, inference demand grows, making efficient serving increasingly important. Discrete diffusion language models (dLLMs) offer a promising path to more efficient inference through parallel token generation. Prior dLLM decoding methods, however, focus on improving inference efficiency for individual sequences. We formulate fixed-budget dLLM batch decoding and introduce Batch-Paced, a training-free method that jointly optimizes the allocation of decoding compute across requests to improve overall batch quality. Across two dLLM backbones, five benchmarks and their representative mixtures, Batch-Paced improves batch performance under matched budgets; gains generally grow with batch size and largely persist across decoding configurations. Analysis of requests' decoding trajectories suggests that these gains arise largely by raising the committed-token confidence of lower-confidence requests. We further show that Batch-Paced's batch decode time remains competitive and that adding a decoding-depth allowance during Batch-Paced decoding can shorten it further at a modest performance cost.

This paper presents a robust control framework for safe navigation of a unicycle robot to a goal among multiple static obstacles under an unknown, time-varying disturbance. The controller solves a single real-time quadratic program (QP) that combines a goal-reaching term with one safety term per obstacle. The goal-reaching term is a cross-track control Lyapunov function built from body-frame tracking errors rather than a bearing angle, which keeps its Lie derivatives bounded everywhere, including the goal. Safety is encoded using robust control barrier functions, one per obstacle, each tightened by a closed-form, state-dependent disturbance bound. When stabilization and safety conflict, a slack variable relaxes only the convergence requirement, so that the safety constraints remain strictly enforced. Whenever the QP is feasible, the robust barrier constraints guarantee forward invariance of every obstacle's safe set under all admissible disturbances. Complementing this safety guarantee, a Lyapunov analysis yields an explicit ultimate bound on the distance to the goal. A learning-augmented variant replaces the fixed objective with a reinforcement-learning reference, which accelerates convergence when multiple constraints are simultaneously active. We further derive the obstacle separation below which the gap between two obstacles becomes too narrow to pass through under the robust constraints. For such gaps, a learned virtual goal for the Lyapunov function steers the robot around the obstacles without compromising safety. Simulations, a Monte Carlo study, and a comparison against a sampled-data robust baseline support the theoretical results.

In stochastic control problems involving optimal portfolio choice, one often seeks an optimizer of some local criterion (dynamic programming value function) generated by some global criterion (e.g., expected utility). The local optimizer then typically acts as a stochastic integrand with respect to some multivariate stochastic process (e.g., asset prices). This paper takes a closer look at local criteria in the form of “local expected utility”. Since the existence of the local optimizer can be established under very mild conditions, it is crucial to understand under what circumstances the local optimizer has good integrability properties. It is shown that the reasonable asymptotic elasticity condition of Kramkov and Schachermayer (1999) applied to the local utility function plays a key role in this context.

Post-hoc linearization is an attractive way to accelerate pretrained Video Diffusion Transformers (Video DiTs), yet existing methods largely treat the quality loss induced by attention replacement as a recovery problem. We show that a more fundamental bottleneck lies upstream: initialization. Random initialization discards pretrained knowledge, whereas manual weight reuse transfers it through rigid and incomplete source–target correspondences. We introduce LinMorph, which recasts Video DiT linearization as a learnable softmax-to-linear weight transformation problem. Instead of copying predefined counterparts, LinMorph learns a cross-parameter morphism that lets each target Gated DeltaNet (GDN) parameter draw jointly from pretrained query, key, value, and output projections, including for parameters with no direct softmax counterpart. A single morphism is shared across layers and learned through local teacher alignment, while producing layer-specific initializations from their respective pretrained weights. The resulting weights are then materialized as independent GDN modules for progressive distillation and joint recovery.Since the morphism operates in weight space rather than token space, it can be learned on short, low-resolution clips and directly reused at larger video scales. Across pretrained Video DiTs, LinMorph enables markedly faster recovery and better final generation quality than random initialization and heuristic weight reuse. Results show that efficient linearization depends not only on how a converted model is recovered, but critically on how pretrained knowledge is transferred before recovery. Project page: https://lonelyling.github.io/LinMorph.

Distributed learning with multiple local steps reduces communication but aggravates drift, and its practical performance often depends on carefully tuned hyperparameters under heterogeneous data and communication constraints. At the same time, recent Muon-type optimizers suggest that constraining updates within suitable norm balls provides a simple way to encode geometric priors into optimization. In this work, we propose NBCLT (Norm-Ball Clipped Local Training), a unified local-training framework for both federated learning (FL) and decentralized learning (DL). NBCLT combines a predict-and-correct local update rule with Euclidean projections onto flexible norm balls, thereby performing norm-ball clipping under flexible update geometries while retaining a simple implementation. In FL, its problem-parameter-free regime achieves $\mathcal{O}(\epsilon^{-2})$ communication complexity and $\mathcal{O}(\epsilon^{-4})$ total stochastic-gradient complexity without assuming bounded heterogeneity, when the per-round local-computation budget is scaled appropriately with the horizon. In DL, NBCLT is compatible with both static and time-varying mixing and preserves these accuracy exponents under the analogous computation scaling, up to fixed network constants and an additive initialization cost.

Lifelong multiagent path planning on construction floors must accommodate continually changing goals and work occupancy. Existing planners can update paths and timing online, but work motion space coupling changes the spatial basis of unfinished encounters, while rerouting affects later encounters. Planning must therefore construct and maintain local encounter possibility. We propose Width based Planning with Compatible Encounter Space Allocation (WP), which constructs shared passage relationships from actual encounter demands and real time widths. Shared channels, ordered anchors, and urgency based updates coordinate successive encounters while retaining individual greedy motion between them. A necessary and sufficient width condition for fixed-point position exchange and a necessary directional condition for composite exchange provide geometric grounds for channel construction and encounter relocation. In crowded floor computational experiments, WP maintained route continuation after occupancy changes and enabled multi agent clearing, with infrequent, brief near stops. Complete run comparisons with four methods showed 25.2% to 52.2% more mean goal arrivals than the best baseline at each configuration of 5 to 10 agents, with lower mean computation time than the rolling horizon K_CBS adaptation. The contribution is a planning principle that constructs and continuously maintains local passage relationships from real time space, unifying shared spatial allocation and conditional local clearing to extend lifelong planning to work motion space coupling.

We prove that network coding and fractional routing have the same capacity region in undirected networks whose demand graph has matching number at most two, that is, when no three demand edges have pairwise disjoint endpoints. The metric-packing reduction of Liu et al. reduces this problem to a distance-weighted communication inequality on the fixed graph $\Gamma_{3,3}$. We first prove two entropy inequalities for sessions between designated terminal pairs on $\Gamma_{3,3}$. We then extend this result to sessions with arbitrary endpoints to obtain the distance-weighted communication bound required by the reduction. The proof combines conditional entropy inequalities with two graph contractions and an information inequality for deterministic interaction. Combining this inequality with the existing reduction establishes routing optimality for this class of demand graphs.

Jumbled scattered factors are a generalisation of scattered factors (or subsequences) where a jumbled scattered factor is obtained from another word by deleting and rearranging letters. For limiting the number of rearrangements, we distinguish $\ell$-jumbled scattered factors where $\ell$ limits the number of letters that may occur rearranged. For instance, $\mathtt{car}$, $\mathtt{start}$, $\mathtt{rat}, \mathtt{cat}$ are jumbled scattered factors of $\mathtt{abstract}$ even though $\mathtt{car}$, $\mathtt{start}$ and $\mathtt{cat}$ are no scattered factors and each need one jumble. An example for a jumbled scattered factor with $2$ jumbles is $\mathtt{tatar}$. In this work, we investigate two problems on jumbled scattered factors from an algorithmic perspective. First, we give algorithms for testing jumbled Simon's equivalence where two words are equivalent if they share the same $\ell$-jumbled scattered factors up to length $k$. Second, we investigate how to count the possible occurrences of a jumbled scattered factor – the jumbled binomial coefficient. After defining the jumbled binomial coefficient, we study jumbled embeddings and factorisations of such as well.

We study the geometry and resulting properties of Fréchet functions and barycenters in metric spaces with curvature bounded above by a positive constant in the sense of Alexandrov. Existing results typically rely on the assumption that the support is small enough that the squared distance, and thus the Fréchet function, is strongly geodesically convex. We work in the sharper regime where existence and uniqueness of the barycenter are guaranteed, but where convexity typically fails. We develop new geometric properties of these spaces, including a quantitative four-point inequality and a geodesic stability result. Using these tools, we establish new finite sample error bounds for two statistical estimators: the empirical barycenter and the iterated barycenter, as well as new results on sparse representations and approximations of barycenters.

In manga translation, visual context outside speech-bubble text, including speaker identities, referents, and scene tone, is depicted in page images, making it natural to refer to images during translation evaluation. However, most conventional manga translation evaluations rely solely on text-only automatic metrics, leaving the impact of image input insufficiently explored. We propose a practical validation framework that employs the same vision-language model (VLM) as both a text-only judge (TOJ) and an image-aware judge (IAJ) to investigate evaluation divergences associated with image-aware evaluation procedures. We use a Japanese-English manga dataset of official translations aligned at the page and speech-bubble levels. First, adding page images to TOJ improved the primary VLM's detection of deliberately injected, visually dependent degradations. Second, we analyzed evaluation divergences between TOJ and IAJ, finding different error counts and error-category distributions. Human verification did not reveal a consistent preference between TOJ and IAJ.

The code generation capabilities of large language models (LLMs) can equally be misused to implement harmful functionality, underscoring the need to understand their safety behavior in response to programming requests. While request phrasing is known to influence refusal, it remains unclear whether the form in which coding task requirements are stated affects the refusal process for malicious code generation. We introduce PseudoSpec, a template that organizes functional requirements as engineering specifications, and investigate its effects through controlled comparisons, representation analysis, and activation interventions. Across five open-weight models, removing explicit risk keywords only partially reduces refusal, whereas specification prompting substantially lowers refusal even when these keywords are retained. This reduction accompanies consistent shifts toward code-output-related representations, and interventions along condition-difference directions alter refusal behavior. Output analysis further distinguishes code-related responses from generation degeneration. Building on these findings, we propose SpecCloak, a black-box red-teaming framework combining PseudoSpec with discriminator-guided optimization, achieving an average execution-verified attack success rate of 28.91% across eight target models on RedCode-Gen. We also introduce SpecReveal, a defense prototype combining natural-language intent reconstruction with dual-view detection without target-model fine-tuning. It achieves a 96.88% attack detection rate in a separate prompt-detection evaluation, with false positives assessed on benign programming and specification-style inputs. These results connect specification-induced response differences to practical security consequences and identify intent reconstruction as a complementary detection approach.

Demand for wireless communication is growing, and commercial satellite services are expanding with it, increasing pressure on a finite radio spectrum. Allocation efficiency directly limits how many users a satellite can serve. We study how to assign frequency sub-channels to the carriers of a multi-beam satellite, maximising the total priority of admitted carriers while keeping each within its interference tolerance. Earlier work on this interference-limited channel allocation problem placed each carrier in a single continuous frequency block. Removing this restriction can admit more carriers, but a direct integer programming model cannot solve the largest test problems in practical time, so new methods are needed. We prove the problem computationally hard, reformulate it around how carriers share sub-channels, and derive three bounds on the optimum. We develop heuristics, including matheuristics combining local search with optimisation models, and an orchestrated portfolio that schedules them, recombines their solutions and stops early once its result is proven optimal or no longer improves. On 24 problems with up to 500 carriers, the best heuristics come within about 1% of the best known solution on average. The portfolio matches or exceeds the best heuristic for each problem on 23 of 24 with equal weights and 22 with priority weights, and is up to about 5% better on larger, more congested problems. It requires about a quarter of the time to run every heuristic and keep the best, and under half that of its heuristics run in a fixed order, which return slightly worse solutions on average.

In this paper, we formulate a novel form of conditional layout generation for graphic design, termed content-edited layout generation(CELG). CELG extends conventional content-aware layout generation (CLG), which generates a foreground layout for a fixed background image. The key distinction is that CELG treats the input image as an editable visual asset and, conditioned on a textual design intention and a foreground requirement that may be empty, produces an adapted background together with its layout. To study this task, we introduce Search-Compare-Edit(SCE), a data construction framework that supplies the paired supervision missing from existing design datasets. Starting from professional design canvases and their layouts, SCE synthesizes plausible pre-edit source images and corresponding editing instructions, yielding InstructEdit, a dataset of aligned source images, target canvases, instructions, and layouts. Based on the generated data, we train a two-flow multimodal flow matching model that couples a pretrained image editor with a layout flow transformer, enabling background editing and layout generation simultaneously within a unified framework that more closely reflects real-world design workflows. Our experiments systematically demonstrate the value of CELG as a distinct task and the performance advantages of the proposed model.

Continual learning (CL) adapts pretrained models to a sequence of tasks while retaining knowledge acquired from earlier tasks. When task data are sensitive, differential privacy (DP) provides record-level protection, but maintaining both adaptation and retention under privacy constraints remains challenging. We study this problem in continual low-rank adaptation (LoRA) and observe that: 1) private training increases forgetting primarily during subsequent adaptation rather than initial task acquisition; 2) jointly perturbing both LoRA factors introduces a non-negligible bilinear noise interaction; and 3) DP aggravates the stability-plasticity dilemma, hindering both knowledge retention and adaptation to new tasks. Based on these observations, we propose DP-SPA, which constructs a stable subspace from previously privatized updates and a plastic subspace from pretrained weights within its orthogonal complement. The method fixes these bases and privately optimizes only the LoRA factor, eliminating the bilinear noise interaction within each step. Before merging each task update, it rescales stable-direction coefficients using historical and current update energies. We establish a privacy guarantee and show that recalibration does not increase historical overlap and tightens a conditional upper bound on old-task loss increase. Experiments across five datasets, eight settings, and multiple privacy budgets show that DP-SPA improves average accuracy and reduces forgetting compared with the evaluated baselines.

Vision-Language Models (VLMs) and world models have demonstrated strong capabilities in semantic understanding and future scene evolution for autonomous driving. However, effectively integrating both capabilities for trajectory generation remains challenging. In this work, we propose DriveWLA, a unified World-Language-Action model based on Mixture-of-Transformers (MoT) that jointly models semantic understanding, future scene dynamics, and trajectory generation. DriveWLA consists of three experts: a World Expert for future prediction, a Language Expert for semantic understanding, and an Action Expert for trajectory generation. We further introduce Cross-Expert Attention, which enables controlled information flow across experts while aligning semantic and dynamic representations. To improve trajectory generation beyond imitation learning, we apply Flow-GRPO reinforcement learning post-training to refine the Action Expert. Extensive experiments show that DriveWLA achieves superior performance in both open-loop and closed-loop evaluation. DriveWLA achieves 92.3 PDMS on NAVSIM v1 and 90.1 EPDMS on NAVSIM v2, demonstrating strong performance among VLM-based and world-model-based planners. Without specific fine-tuning, DriveWLA further demonstrates strong zero-shot generalization to HUGSIM, nuScenes, and Bench2Drive. These results validate the effectiveness of our unified World-Language-Action modeling, establishing DriveWLA as a generalizable framework for autonomous driving.

Visual question answering models often produce plausible responses even when the available visual evidence is insufficient to support a valid answer. Although prior work has introduced benchmarks and abstention mechanisms for unanswerable visual questions, it remains unclear which signals derived from a frozen vision-language model reliably indicate answerability on unseen image-question pairs. We systematically evaluate answerability signals on a stratified 10K subset of UNK-VQA using frozen InstructBLIP, covering predictive uncertainty, cross-modal alignment, visual grounding, visual dependence, internal representations, and learned probes. Answerability discrimination remains weak, with test AUROC largely confined to 0.50-0.54. CLIP alignment yields the highest point estimate but does not significantly outperform generation confidence under paired bootstrap analysis. Targeted removal of question-relevant visual evidence causes substantially larger changes in answer confidence than same-size random masking, yet intervention sensitivity provides essentially no answerability discrimination. Increasing probe capacity, combining multiple signals, and adding reason-aware supervision do not provide robust improvements, while post-freeze cross-validation yields similarly weak discrimination. Selective-answering analysis shows limited usable coverage under low-risk operating criteria. Failure-mode analysis shows that the relative utility of individual signals varies across perturbation mechanisms. Taken together, these results show that, on our UNK-VQA subset with frozen InstructBLIP, visual dependence, uncertainty, semantic alignment, and static representations are not robust standalone proxies for answerability. Our findings emphasize the need to distinguish a model's reliance on visual evidence from whether that evidence is sufficient to support an answer.

Long-context training of large language and multimodal models couples structured attention masks (document, prefix-LM, share-question, sliding-window, block-sparse) with multi-node context parallelism, making the mask a dominant bottleneck. FlashMask's column-wise sparse mask compresses them from $O(n^2)$ to $O(n)$ via four query-axis boundary vectors, covering broad structured patterns. Implemented for Ampere GPUs, it neither exploits FlashAttention-3's TMA/WGMMA warp-specialized pipeline on Hopper nor supports distribution. The column-wise sparse mask is indexed by key positions but stores query-axis interval boundaries. Under any multi-chunk context-parallel assignment, it reduces per-rank mask localization to a provably zero-communication clip–shift–max on the boundary vectors, compatible with the single-GPU kernel. FlashMask-3 introduces a mask-aware Hopper kernel with three-role warp specialization atop FlashAttention-3's TMA/WGMMA pipeline, plus a mask-aware distributed extension: zero-communication mask localization, a sparsity-aware load balancer (HCC-IPO), and a mask-aware compute-communication overlap with mask-driven communication pruning, whose forward all-gather uses topology-aware hierarchical routing. On a single H100 across 12 masks at 4K–128K context, it improves achieved TFLOPs over FlashMask, FlexAttention, and MagiAttention by 45–141%, 34–71%, and 2–44%. At context-parallel degrees 4–32 with 8K sequence per device, it attains 31–75% compute-communication overlap efficiency and improves throughput over Megatron-Core CP ring, Megatron-Core CP all-gather, and MagiAttention by $1.80\times$–$3.71\times$, $1.71\times$–$2.92\times$, and $1.09\times$–$1.46\times$. In end-to-end training of a 118B-A14B MoE at 128K on 128 H100 GPUs, it raises throughput by 7.3–11.0% over a Dual Chunk all-gather baseline, with attention forward+backward 24.5–44.9% faster.

Extended chain-of-thought reasoning improves problem-solving performance but increases inference cost and latency, motivating more accurate answers with shorter reasoning. We introduce PaceReasoner, a label-free test-time reinforcement learning framework that uses a model's LONG responses to improve its SHORT-mode accuracy. Both modes share a trainable policy, creating a need for reliable supervision, sufficient SHORT training samples, and continued LONG responses to guide learning. PaceReasoner combines all-response and LONG-only answer consensus through progressive, confidence-weighted supervision. A behavior reward gradually shifts from encouraging SHORT exploration to rewarding agreement with LONG answers, while a ratio controller regulates the balance between modes. Across three model backbones and four benchmarks, PaceReasoner improves average SHORT accuracy over the baselines by 7.4–13.5 percentage points, with fewer than 25 reasoning tokens per response on average. Ablations support the key components, and cross-dataset results demonstrate transfer beyond the adaptation set.

Virtualization enables mixed-criticality applications to run on shared hardware by isolating them within separate virtual machines (VMs). However, achieving end-to-end timing guarantees for critical tasks hosted on virtual CPUs (VCPUs) without relying on static partitioning is hard. A possible solution is a time-triggered hypervisor that dispatches VMs' VCPUs to processor cores according to a predetermined global schedule. A global schedule includes the computation, the network, and the virtualization layer to provide end-to-end timing guarantees across hosts and VMs during design time. In this work, we propose a VirtIO-based virtual network that realizes IEEE 802.1Qbv quasi-compliant inter-VM communication, providing predictable timing for critical messages exchanged between time-triggered VMs. We present our VirtIO-based virtual network architecture and derive network-level scheduling constraints that emerge from the properties of the proposed solution. We implement the proposed design in the ACRN hypervisor and evaluate the end-to-end latency of critical traffic in a loaded network, demonstrating that our time-triggered VirtIO-net prototype provides predictable end-to-end latency under load.

Sparse autoencoders decompose language model representations into interpretable features, but feature interpretation in this literature remains predominantly correlational. Attribution methods offer more direct estimates of feature influence, yet their faithfulness to actual causal effects has not been systematically evaluated in the SAE setting. We apply SHAP-based attribution to SAE latent features and evaluate it against direct causal intervention, using sentiment classification on SST-2 with GPT-2 Small and a 32,768-feature layer-7 SAE as a testbed. We compare SHAP to Integrated Gradients, Gradient$\times$Input, and probe-weight baselines across linear and MLP probe arms, scoring each against feature ablation and activation steering measured through the language model's own logit difference. Most faithfulness associations we measure do not survive multiple-comparison correction, and attribution methods that agree strongly with one another ($\rho = 0.42$–$0.80$) agree poorly with measured causal effects, indicating that inter-method agreement is not evidence of causal faithfulness. Two effects hold under control. First, local per-example attribution retains a directional advantage over global averaged attribution, while its apparent importance advantage is explained by feature activity rather than superior causal ranking. Second, under known-ground-truth minimal pairs, attribution recovers inserted sentiment signal well above chance (AUROC $0.792$) but fails to localize deletions (AUROC $0.520$), suggesting that these methods detect the presence of a feature's influence more reliably than its absence. We argue that importance faithfulness and directional faithfulness are separable properties requiring separate evaluation, and that attribution agreement should not substitute for causal validation.

Does there exist an 8-variable balanced Boolean function with nonlinearity 118? This is a famous and long-standing open problem in Boolean functions. In this paper, we give a negative answer to this open problem, and prove that the maximum nonlinearity of balanced Boolean functions in 8 variables is 116, establishing the 8-variable instance of Dobbertin's conjectured recursion for the minimum Walsh peak of balanced functions. Equivalently, the covering radius of the Reed–Muller code RM(1,8) in 0-resilient Boolean functions is 116. We deduce the result by obtaining a quadratic form of polar rank 6, and reducing the problem to the existence of a 6-variable Boolean function with nonlinearity greater than 26 whose Hamming distance to a bent function is between 4 and 6.

Rainfall forecasts over the Western Himalaya remain poor, and the relative contributions of the cumulus and microphysics parameterizations to that error are not well constrained.

Coding agents are tasked with producing sophisticated interactive applications such as websites, desktop applications, and games. While humans assess these applications interactively, current automated LLM evaluations primarily focus on code analysis and unit tests. This mismatch can cause LLM judges to miss failures that are immediately apparent to human users. We study Interaction-as-a-Judge, an evaluation paradigm where computer-use agents (CUAs) directly interact with an application. We demonstrate that Interaction-as-a-Judge can improve both evaluation and generation of interactive applications. To this end, we collect and release two complementary datasets: $6,858$ pairs of applications with human preferences and corresponding agent traces, and $222$ applications with human-verified requirement labels. Relative to standard non-interactive approaches, we find that Interaction-as-a-Judge improves evaluation in terms of predicting human preferences ($65.6\% \to 69.1\%$) and requirement verification ($77.0\% \to 81.6\%$). We also observe scaling benefits: across both preference alignment and requirement verification, higher-quality interactions improve performance. Finally, through both inference- and training-time experiments, we demonstrate that Interaction-as-a-Judge can also improve the quality of generated artifacts. At inference time, Interaction-as-a-Judge repairs $75.5\%$ of broken requirements within five attempts, compared to $55.4\%$ for code-only self-feedback. At training time, using Interaction-as-a-Judge as a reward signal improves win-rate over the base model by $6.7$pp while reducing both generated code length by $35.7\%$ and build failure rates by $7.2$pp compared to a code-only reward signal.

We present Swiss4Arg, a multilingual dataset for argument mining based on official Swiss voting brochures, covering all four national languages: German, French, Italian, and Romansh (a Romance minority language exclusively spoken in Switzerland). The brochures contain explicitly formulated arguments systematically organized into pro and con sections, resulting in a cross-lingually aligned resource with balanced argumentative stance. We create a dataset that comprises 53 federal voting issues collected between 2019 and 2025, annotated for argument components (major claims, claims, premises, and non-argumentative units) and relations (support and attack) by three native speakers per language. Inter-annotator agreement (Krippendorff's alpha) reaches 0.548 for component classification and 0.623 for relation and target classification, computed jointly across all annotators on the parallel data. We release both adjudicated labels and individual annotator annotations, enabling analyses of disagreement and cross-linguistic variation, and provide baseline results using GPT-5.4 and SwissBERT. Swiss4Arg establishes a novel benchmark for multilingual argument mining, with a particular focus on cross-lingual comparability and low-resource settings such as Romansh. The dataset and code are publicly available at https://anonymous-hf.up.railway.app/a/gozw0nion72y/ (HuggingFace) and https://anonymous.4open.science/r/MultiVote-03BD/ (GitHub).